Changeset 802

Show
Ignore:
Timestamp:
10/10/08 14:49:46 (17 months ago)
Author:
arsalan
Message:

Edits

Location:
docs/Lowthane
Files:
5 modified

Legend:

Unmodified
Added
Removed
  • docs/Lowthane/design.tex

    r799 r802  
    150150\begin{figure} [h] 
    151151\centering 
    152 \includegraphics[width=0.8\columnwidth]{figs/topology_collection.png} 
     152\includegraphics[width=0.5\columnwidth]{figs/topology_collection.png} 
    153153\caption{An example collected topology when only one link is reported by each 
    154154  node.  We report four links per node.}\label{fig:topology} 
     
    262262\begin{figure} [h] 
    263263\centering 
    264 \includegraphics[width=0.8\columnwidth]{figs/routing.png} 
     264\includegraphics[width=0.5\columnwidth]{figs/routing.png} 
    265265\caption{The effect of the route install primitive.}\label{fig:routing} 
    266266\end{figure} 
     
    322322%% cause thrashing, which leads to degraded performance.  We explore this in 
    323323%% detail under various conditions and workloads in section~\ref{sec:eval}. 
    324 \subsection{Forwarding Discipline} 
     324\section{Our Hybrid Approach} 
     325 
    325326{\bf Decision:} The default route as used as a backup in the case where installed state 
    326327  is invalid. 
     
    427428\subsection{Shortcomings of Centralized Protocols} 
    428429 
    429 Introduction of a centralized element of routing raises a family of concerns.  We briefly examine the most common ones raised in the literature and anecdotally in the context of our setting and approach.\\  
    430 \\ 
     430Introduction of a centralized element of routing raises a family of concerns.  We briefly examine the most common ones raised in the literature and anecdotally in the context of our setting and approach. 
    431431\noindent 
    432432{\bf Latency of sending packet through controller: }The latency incurred by 
     
    437437real-time scenarios in which latency is critical; in such cases an on-demand 
    438438solution like {\lowthane} may not be appropriate without quality-of-service 
    439 adaptations.\\  
    440 \\ 
     439adaptations. 
     440 
    441441\noindent 
    442442{\bf Controller becomes a single point of failure: }The failure of the 
     
    448448in the network and the number of controllers.  We discuss this further in 
    449449section~\ref{sec:extensions}, but for simplicity focus on single-controller 
    450 networks until then.\\  
    451 \\ 
     450networks until then. 
     451 
    452452\noindent  
    453453{\bf A persistent back channel to the controller is needed: }The directed acyclic 
     
    459459to dynamic topological conditions, and also maintains multiple options for 
    460460additional reliability.  Our results in section~\ref{sec:eval} demonstrate the 
    461 reliability of this back channel.\\ 
    462 \\ 
     461reliability of this back channel. 
     462 
    463463\noindent 
    464464{\bf Difficult to maintain consistent global view of topology: }This 
     
    473473initial evidence that it is possible to react sufficiently quickly to network 
    474474changes.  
    475 \\  
    476 \\ 
     475  
    477476 
    478477\subsection{ROLL Requirements} 
  • docs/Lowthane/evaluation.tex

    r796 r802  
    3131\begin{figure} [h] 
    3232\centering 
    33 \includegraphics[width=0.8\columnwidth]{figs/Soda-4_2008-09-24.png} 
     33\includegraphics[width=0.5\columnwidth]{figs/Soda-4_2008-09-24.png} 
    3434\caption{79-node Smote Testbed at UC Berkeley}\label{fig:smote} 
    3535\end{figure} 
  • docs/Lowthane/lowthane.tex

    r796 r802  
    3939high control overhead, or poor response to loss.  We propose a new design for 
    4040routing in this class of networks which takes advantage of typical device 
    41 heterogeneity and traffic patterns in this environement.  We present a hybrid 
     41heterogeneity and traffic patterns in this environment.  We present a hybrid 
    4242between a centralized and distributed routing algorithm, which uses a 
    4343distributed algorithm to provide reliable routes to a small set of 
    4444controllers.  Given that the predominant traffic patterns in these networks  
    4545are collection and dissemination, these routes optimize for the common case. 
    46 The focus of this work is to evalute the potential for further optimization of 
     46The focus of this work is to evaluate the potential for further optimization of 
    4747point-to-point traffic within the network through the use of a centralized 
    4848solution.  In our design, controllers maintain a view of the network topology as 
     
    5555performance, but to present preliminary results showing that it performs well 
    5656in a range of expected situations, thus demonstrating the design's viability. 
    57 Finally, we evalute the state and traffic complexity of our design to 
     57Finally, we evaluate the state and traffic complexity of our design to 
    5858demonstrate that it meets requirements for the space. 
    5959 
  • docs/Lowthane/related.tex

    r797 r802  
    3737single destination routing, and also affords no flexibility for dynamic 
    3838topology conditions (unless links completely break).  TSMP, which is 
    39 incorporated in the WirelessHART standard is a well-established industry 
     39incorporated in the WirelessHART standard, is a well-established industry 
    4040approach that includes slotted time \& channel scheduling using the global 
    4141information, but little technical data is available for it. 
     
    6565 
    6666Finally Hui's work explores extending the 
    67 internet architecture to L2Ns, and develops the algorithms necessary for 
     67Internet architecture to L2Ns, and develops the algorithms necessary for 
    6868default route formation in \cite{hui}.  This work has been invaluable in 
    6969helping us to frame our architecture. 
  • docs/Lowthane/tossim.tex

    r801 r802  
    3737\begin{figure}[h] 
    3838\centering 
    39   \includegraphics[width=.8\columnwidth]{figs/collection.pdf} 
     39  \includegraphics[width=.5\columnwidth]{figs/collection.pdf} 
    4040\caption{Our default route selection protocol compared with CTP in simulation 
    4141  with a poor noise model, and to results on our testbed.}